† Graduate School of Public and International Affairs and Institute of the Environment, University of Ottawa, 55 Laurier Avenue East, Ottawa, Ontario, K1N 6N5, [email protected]. ‡ Department of Economics and Institute of the Environment, University of Ottawa, 55 Laurier Avenue East, Ottawa, Ontario, K1N 6N5, [email protected].
Salience of Carbon Taxes in the Gasoline Market
Nicholas Rivers† Brandon Schaufele‡
June 10, 2013
Abstract We demonstrate that the carbon tax imposed by the Canadian province of British Columbia caused a decline in short-run gasoline demand that is significantly greater than would be expected from an equivalent increase in the market price of gasoline. That the carbon tax is more salient, or yields a larger change in demand than equivalent market price movements, is robust to a range of specifications. Along with calculating the reduction in carbon dioxide emissions attributable to the tax, we discuss potential explanations for the differential consumer responses to the carbon tax relative to the market-determined price.
Keywords: Carbon tax, tax salience, environmental pricing, gasoline demand
JEL Codes: C26, H23, H29, Q41, Q58
1
1. INTRODUCTION
On July 1st, 2008, the Canadian province of British Columbia (BC) enacted North
America's first broad-based carbon tax designed to reduce greenhouse gas emissions. While
several jurisdictions have implemented emission reduction programs, no other state or province
has implemented a policy that is as ambitious and comprehensive as the BC policy, and no other
North American greenhouse gas control policy taxes households directly based on emissions.1
Media has both lauded the BC carbon tax as a major policy achievement and condemned it as a
“nightmare” for industry and families.
While carbon taxes have been in place in various European countries since the 1990s,
econometric analysis of their impact is limited,2 and there is minimal evidence on the
effectiveness of similar programs in the North American context. Research on the implications
of an actual carbon price is particularly important as several states and provinces are currently
debating whether or not to introduce similar broad-based policies.3 By exploiting cross-
provincial panel data variation, a range of robustness checks as well as instrumental variables
derived from the policy’s design, this study provides the first causal evidence of the effect of a
carbon tax on the short-run gasoline decisions of North American households.
Throughout our analysis we concentrate on the saliency of the BC carbon tax. Tax
saliency refers to the hypothesis that tax-induced price changes generate distinct demand
responses when compared with equivalent market-determined price movements. An emerging
1 The Regional Greenhouse Gas Initiative (RGGI) is a leading example of a US emission control program. A joint initiative of nine states – Connecticut, Delaware, Maine, Maryland, Massachusetts, New Hampshire, New York, Rhode Island, and Vermont – its objective is to develop a market‐based program aimed at “reduc[ing] CO2
emissions from the power sector [by] 10 percent by 2018” (RGGI, 2012). Similarly, the Canadian provinces of Alberta and Quebec have enacted carbon pricing policies. Alberta effectively taxes large industrial emitters (>100,000 tonnes) at $15 per tonne of CO2, while Quebec has a carbon tax on natural gas, coal and petroleum equal to $3 per tonne CO2. 2 Anderson (2010) describes European carbon taxes, and provides some evidence on their effectiveness using a simulation model. Econometric research on carbon taxes includes Martin, de Preux and Wagner (2011) who examine the UK’s Climate Change Levy and Enevoldson, Ryelund and Andersen (2007) who examine Scandanavian industrial carbon taxes. It is difficult to use the limited empirical evidence from these studies in other contexts because (1) European carbon taxes were often introduced as replacements for existing energy taxes, not as stand‐alone taxes, (2) they include numerous exemptions and differences in rates across sectors (Bruvoll and Larsen, 2004; Sumner, Bird and Smith, 2009), and (3) there may be different preferences, culture, or geography in these countries compared to others. 3 Most notable is the state of California which adopted a state‐administered cap‐and‐trade system on October 20, 2011. California’s emission trading market is expected to be the cornerstone of the larger Western Climate Initiative, a collaborative effort with four Canadian provinces to develop a North American greenhouse gas emissions trading program.
2
literature delves into this hypothesis (see Chetty, Looney and Kroft (2009), Finkelstein (2009),
Congdon, Kling and Mullainathan (2009), Goldin and Homonoff (2010) and Gallagher and
Muehlegger (2011)). This paper, in particular, can be viewed as an extension of Davis and
Kilian (2010) and Li, Linn and Muehlegger (2012), research which examines responses to excise
taxes relative to price changes induced by supply shocks in the gasoline market. By evaluating a
situation where a carbon tax actually exists however, this study makes an important contribution
to this growing literature. To the best of our knowledge, ours is the first empirical investigation
into the saliency of environmental taxation. Carbon taxes differ from excise or other
consumption taxes in that, by imposing a disincentive on fossil fuel consumption, they are
explicitly designed to reduce environmental externalities. Even though excise and sales taxes
reduce gasoline demand in the short-run, they are not overtly designed to correct environmental
externalities. Revenues from gasoline taxes, for example, are frequently earmarked for road
infrastructure, projects which lower the long-run costs of driving. Concentrating on carbon
pricing permits us to identify the relative saliency of a carbon tax compared with the underlying
market price of gasoline when the unambiguous purpose of the policy is to reduce gasoline
demand.
Our main result is that the BC carbon tax generated demand response that is 7.1 times
larger than is attributable to an equivalent change in the carbon tax-exclusive price. In our
preferred model, a five cent increase in the market price of gasoline yields a 1.8% reduction in
the number of litres of gasoline consumed in the short-run, while a five cent increase in the
carbon tax, a level approximately equal to a carbon price of $25 per tonne, generates a 12.5%
short-run reduction in gasoline demand. These results lead us to claim that the carbon tax is
more salient than market-determined price changes: carbon taxes produce larger demand
responses than tax-exclusive price increases. Due to the robustness and consistency of our
estimates across a range of specifications including models that incorporate intuitively appealing
and strong instrumental variables, we feel that our results can be interpreted as causal. We
believe that it was BC’s carbon tax that caused the decline in provincial gasoline demand. We
also use our econometric results to construct counterfactual scenarios in order to calculate the
change in gasoline-related emissions stemming from the carbon tax. We find that the BC policy
reduced carbon dioxide emissions from gasoline consumption by more than 3.5 million tonnes.
Of this total, 85.6%, or 3.1 million tonnes, is due to the additional saliency of the carbon tax –
3
i.e., it is an environmental bonus that would not have been achieved if individuals responded to
carbon taxes in the same way as to identical changes in gasoline prices caused by other factors.
Our results are in line with Li, Linn and Muehlegger (2012) who find that consumers are
more responsive to changes in gasoline excise taxes than to tax-exclusive prices. In particular,
Li, Linn and Muehlegger estimate a tax saliency ratio (i.e., the mean consumer response to an
increase in gasoline taxes divided by an equivalent increase in market prices) equal to 8.1, a
value that is within the range of our estimates. Although in a different context, our results also
complement Finkelstein (2009) and Chetty, Looney and Kroft (2009) who suggest that
consumers exhibit more elastic demand response if prices increase due to a highly visible tax
than if prices increase for some other reason. Finkelstein shows that the demand curve for
driving becomes more inelastic when tolls are charged electronically as compared to manual
collection. Chetty, Looney and Kroft use experimental evidence to demonstrate that making a
sales tax visible increases demand responsiveness. Yet it is not obvious that the retail gasoline
market is sufficiently similar to electronic tolls or grocery store purchases to enable direct
comparisons. Unlike most goods, gasoline prices are advertised as tax-inclusive. Both excise
and sales taxes are included in stated rates, so buyers pay exactly the price that they see on signs
and at the pumps (i.e., there is no risk of making mathematical errors between the instant
consumers decide to buy fuel and the point where they must pay for the purchase).
The remainder of this paper contains five sections. Section two describes the design of
the BC carbon tax policy. Section three presents our data and empirical methodology. The main
results are in section four. Section five discusses the results including potential explanations for
the added saliency of the carbon, calculations of the reduction of carbon dioxide emitted and
possibly confounding policies that are coincident with the carbon tax. Section six concludes.
2. DESIGN OF THE BC CARBON TAX
The announcement that BC was introducing a carbon tax came as a surprise to the vast
majority of residents (Harrison, 2012). The province’s Finance Minister formally revealed the
revenue-neutral carbon tax in her February 2008 budget speech. By July 1st, 2008, BC became
the first jurisdiction in North America to have a significant carbon tax on all fossil fuels
4
purchased within its borders.4 It was only during the second half of 2007 that the government
began to hint that environmental pricing was possible. Even then, there was no public
acknowledgement that carbon taxes were a prospective policy option until a speech late in
October 2007.
There are two prominent features of the BC carbon tax design. First, it was implemented
gradually. The tax came into effect on July 1st, 2008. Initially set at $10 per tonne carbon
dioxide equivalent (tCO2e), the tax increased by $5/tCO2e each July 1st until 2012, when it
reached its current level of $30/tCO2e. The $10/tCO2e tax implied a 2.34 cent increase in the
price of a litre of gasoline. At $30tCO2e, this represents an increase in the price of gasoline
equivalent to 6.67 cents per litre. Table 1 presents the progression of the carbon tax in tCO2e and
cents per litre of gasoline. Gradually increasing the tax was intended to minimize potential
adjustment costs associated with the tax shift.
The BC carbon tax was also designed to be revenue-neutral. Revenue-neutrality meant
returning all carbon tax revenues to residents via adjustments to personal and corporate taxes as
well as lump-sum transfers. Several components of the personal and corporate tax schedule were
adjusted to offset the revenues generated by the carbon tax. These changes are illustrated in
Table 1. First, the BC government lowered rate of tax on the bottom two personal income tax
brackets. For a household earning a nominal income of $100,000, Table 1 shows that the
average provincial tax rate was reduced from 8.74% in 2007 to 8.02% in 2008. Two lump-sum
transfers were also included to protect low-income and rural households. Low income
households receive quarterly rebates, which, for a family of four, equal approximately $300 per
year, and beginning in 2011, northern and rural homeowners received a further benefit of up to
$200. Finally, taxes on corporations and small businesses were reduced. BC has two corporate
tax rates, a high and low income rate. High income firms are those that earn profits above the
provincial business limit. In conjunction with the carbon tax, this limit was increased from
$400,000 to $500,000 in 2010. Following the introduction of the carbon tax, the BC government
cut both corporate and small business taxes. The corporate income tax rate was reduced in three
steps. It went from 12.0% to 11.0% in 2008, from 11.0% to 10.5% in 2010 and was finally 4 The carbon tax applies to all emissions that result from the burning of fossil fuels. These account for 77% of total provincial emissions. The remaining 23% is not due to the combustion of fossil fuels, but includes: 10% from non‐energy agriculture and landfills, 9% from fugitive emissions, and 4% from non‐combustion industrial process emissions (BC Ministry of Finance, 2012a).
5
reduced to 10.0% in 2011. Similarly, Table 1 shows that the small business tax rate, which is
applied to profits up to the provincial business limit, was cut by 1.0% in both 2008 and 2009.
The revenue-neutrality of the carbon tax yields several benefits. First, it likely increased
public acceptability for the policy. Since residents’ tax burden did not increase, government was
able to promote the policy as a “tax shift” rather than a tax increase. Second, by cutting personal
and corporate income taxes (yet still keeping total government revenues constant), the policy
minimizes the negative economic impacts of taxation and lowers the marginal cost of public
funds. Extensive theoretical analysis of environmental tax shifts indicates that, in some cases,
economic activity might even increase as a result (Goulder, 1995). As a final point, revenue-
neutrality acts as a commitment mechanism for the government, particularly as the carbon tax
was implemented as the economy entered a recession. Once personal and corporate taxes had
been reduced, the BC government needed the revenues generated by the carbon tax which
equalled $741 million in 2011 (BC Ministry of Finance, 2011). Discussions on delaying the
scheduled increases had to be weighed against deficits and reduced social program spending.
The revenue-neutrality of the BC carbon tax design also provides a potential
identification strategy. By construction, increases in the carbon tax are inversely related to
personal and corporate tax revenue. This unique feature of the policy allows us to formulate
economically sensible instrumental variables described in the empirical methodology section.
While our analysis focuses on the BC carbon tax, we also include – but do not emphasize
– the much smaller Quebec carbon tax in our data and analysis. Quebec introduced a carbon tax
in October 2007. The stated goal of the tax is to raise revenue to pursue environmental projects,
in contrast to the BC tax which is aimed at discouraging fossil fuel combustion. Quebec’s goal is
to annually raise $100 million from the tax; as a result, the tax rate is adjusted each year based on
projected fossil fuel consumption. In practice the tax rate has remained relatively constant at
about $3/tCO2e since it was introduced.
3. EMPIRICAL METHODOLOGY
3.1 Data5
5 Data and code written in R for all of the empirical models will be made available on the authors’ websites in addition to complying with the journal’s data availability policy.
6
Our empirical analysis requires a dataset assembled from several sources. Statistics
Canada (2012) collects aggregate monthly data on litres of premium, mid-grade and regular
gasoline sold within each Canadian province.6 As in Small and Van Dender (2007), aggregate
provincial sales are divided by population aged 15 years or older yielding a variable on a per-
adult basis. Price and excise tax information is retrieved from Kent Marketing Services Limited
(2012). All prices and taxes are inflation-adjusted using monthly, province-specific consumer
price indices. The retail price in each province’s largest urban centre proxies for the price for the
entire province – while a high degree of price correlation within provinces exists, this may mask
some intra-provincial heterogeneity. Interprovincial variation is notably greater than within
province deviations largely due to differential tax levels. Appendix Table A1 presents summary
statistics for the monthly gasoline consumption and tax-inclusive prices by province. Additional
data needed for controls, robustness checks and instrumental variable models such as after-tax
income, share of small and compact car sales and income tax rates and revenues are from several
sources including Statistics Canada (2012), DesRosiers Automotive Consultants Inc. (2012) and
the BC Ministry of Finance (2012b). Our period analysis is January 1990 through December
2011 but we perform checks on shorter subsets of this timeframe. Regressions are run primarily
on a monthly basis, yet selected models use annual data. The monthly dataset contains 2580
province-month observations. The annual dataset, which is generated by aggregating per capita
litres consumed over the year and taking a simple per litre averages for prices and taxes, has 216
observations. Annual models supplement our preferred monthly specifications as several control
variables are only available on a yearly basis.
As a first step, we compare the price and gasoline time series for the provinces of BC and
Ontario (Canada’s most populous province). Figure 1 illustrates this comparison. The upper
panels in Figure 1 show the monthly real per litre tax-inclusive price of gasoline, while the
bottom panels display the corresponding level of gasoline sales.7 The introduction of the carbon
tax on July 1, 2008 is shown by the dashed vertical line. Also shown are simple linear time
trends for the period from January 1, 2000 until the introduction of the tax and for the period
6 For the province of Nova Scotia, data do not start until March 1992. Data are unavailable for the province New Brunswick until November 1992. 7 In this figure, the gasoline demand series has been seasonally adjusted by regressing monthly demand on year dummies and month*province dummies, allowing a distinct seasonal pattern in each province. For each province, the month dummies are then subtracted from the gasoline demand series.
7
following the introduction of the tax. For both BC and Ontario, the January 2000 to July 2008
period is characterized by an upward trend in prices and a congruent decrease in quantity
demanded (i.e., demand curves slope downward). During the month when the carbon tax was
implemented however, prices in both provinces fell rapidly and a stark contrast between
consumption patterns in BC and Ontario is obvious. In Ontario, quantity demanded increases
with the price decline. The opposite occurs in BC where despite the reduction in the tax-
inclusive price the demand continues to trend downward. Figure 1 provides strong suggestive
evidence for the salience of the carbon tax, specifically that the carbon tax acted as a demand
shifter in addition to having a direct price effect.
There are several additional points to note about the data. First, since 2004, the highest
per litre prices in Canada are found in BC. We would expect the price gap between BC and the
rest of the country to grow as the scheduled carbon tax increases were enacted. Yet, it is difficult
to visually distinguish a wedge between provinces.8 Indeed, the gap between BC, Ontario and
Quebec has appears to have shrunk. Further, although the degree of interprovincial price
correlation is high, gas prices do vary across the country. As mentioned, these differences reflect
differential gasoline taxes, but also distinct transportation costs and fluctuations in provincial
supply conditions. Lastly, it should be emphasized that a five cent increase in the price of
gasoline, roughly equal to $25tCO2e, is well within one standard deviation for all provinces.
3.2 Model
Econometrically, we follow the approach of Li, Linn and Muehlegger (2012) and
decompose the retail price of gasoline, the price that consumers see at pumps, into two
components: a carbon tax-exclusive price and a carbon price. More precisely, we specify the
following linear equation:
(1)
where yit is the per capita number of litres of gasoline consumed in province i during period t
(months or years), pit is the carbon tax-exclusive inflation-adjusted price of gasoline in province i
during period t,9 τit represents the inflation-adjusted carbon tax in province i during period t, δi is
province-specific fixed effect that captures geographically constant and time invariant 8 Figure available from authors upon request. 9 Prices include both provincial and federal excise taxes.
ittiitititit py θX
8
unobservable characteristics, while γt is a time fixed effect reflecting either month-years or years
(depending on the resolution of the data) and captures time specific unobserved factors that may
influence gasoline demand. Xit represents other potentially relevant variables that vary at the
province-year level. These include provincial after-tax income and the share of small and
compact cars in the vehicle stock. Finally, is a province and period specific error term.
Robustness checks using first-differenced models are also presented.
Our preferred specification is a log-linear model as this provides a straightforward
interpretation of the effects of the carbon tax (Yatchew and No, 2001). The BC carbon tax is
published in cents per litre, thus coefficients, which approximately reflect semi-elasticities or
percent change in gasoline demand for a given level of tax, have intuitive appeal.10 Finally,
while equation (1) separates prices, taxes and geographical fixed effects, many of our models
interact the provincial dummies with prices and taxes respectively (e.g., ).
We find that demand for gasoline in BC is more elastic than for the country as a whole and this
influences our ability to make inferences with respect to the saliency of prices and taxes.
Interactions enable us to identify the price and tax responses separately for the province of BC.
3.3 Identification
Our identification strategy includes three components. First, we exploit the panel
structure of the data. Next, instrumental variables are introduced. Finally, we perform a series
of robustness checks which involve including potentially omitted variables and focusing on
particular sub-samples. These checks attempt to eliminate other explanations for our results.
Only after our key parameters are shown to be robust across a wide-range of specifications do we
interpret the results as causal.
The primary method of identification is derived from the panel structure of the data.
Time-invariant, region-specific characteristics such as geography are accommodated by the
provincial fixed effects. Time-varying but non-province-specific unobservables (e.g., trade
policy) are captured by the time fixed effects. Any remaining province-time varying factors are
grouped into the error term or included in the model. Provided gasoline prices and carbon taxes 10 Throughout this paper, all semi‐elasticities are calculated as the exponent of the coefficient minus one (e.g.,
); however, estimates which are small in absolute value can be interpreted directly as a semi‐elasticity as the difference is negligible.
it
iitiitp
1e
9
are uncorrelated with the error term, least squares yields unbiased and consistent estimates of the
price ( ) and the carbon tax coefficients ( ).
There are persuasive reasons to believe that both the market price of gasoline and the
carbon tax are exogenous in our empirical models. The gasoline market is generally considered
competitive: wholesale gasoline price is determined in West coast market, and gasoline retailing
generally involves small margins. This argument is supported by Marion and Muehlegger (2011)
who demonstrate that under normal market conditions any change in excise taxes is fully
reflected in final prices – i.e., any change in taxes is “fully and immediately passed on to
consumers” (1202). Similarly, the BC carbon tax was introduced within a short period of time,
caught most residents by surprise, comprehensively covers all fossil fuels and the policy’s sole
objective is to reduce greenhouse gas emissions (i.e., not to influence other outcomes), so it is
unlikely to be confounded. Considering this, both the BC carbon tax and the market price of
gasoline are probably exogenous variables. Our preferred models are therefore parsimonious
and use least squares to identify the parameters.
Nonetheless, while convincing arguments support the exogeneity of gasoline prices and
the carbon tax, we want to eliminate doubt regarding potential simultaneity or omitted variable
bias. We do this by exploiting instrumental variables. Conceivable channels exist through
which bias may arise. For instance, recent construction of bicycle-only lanes in some Canadian
cities in a time period coincident with BC’s carbon tax may have caused some drivers to forego
vehicle ownership just as the tax was introduced. This decision reduces congestion and lowers
the cost of driving for the remaining motorists. Lowering the cost of driving then may encourage
some households to drive more. Thus the net effect of the bike lanes on driving may lead us to
either under- or over-state the true impact of the carbon tax (since our models do not include a
bike lane variable). Along the same lines, Davis and Kilian (2010) claim that macroeconomic
factors such as 2008 recession influenced gasoline prices. Inasmuch as the recession has
dissimilar consequences across provinces bias could arise. Although these effects are likely
second-order, we use instrumental variable models estimated via two-stage least squares to
supplement our core results, since well-designed instruments allow us to consistently estimate
key parameters in the presence of omitted variables.
10
As either or both the market price of gasoline or the carbon tax could be endogenous,
instruments are needed for each. Aggregate provincial personal and corporate income tax
revenues instrument for the carbon tax and provincial gasoline excise taxes instrument for
gasoline prices. Monthly data on excise taxes are available. So, we estimate monthly models
where the price is the potentially problematic variable. Income tax information is available
annually. Accordingly, models for which the carbon tax is potentially endogenous are only
estimated using annual data. Two annual formulations are investigated: a model where the
carbon tax is potentially problematic and a model where both carbon tax and the price of
gasoline are deemed to be potentially endogenous.
Two-stage least squares enables the estimation of unbiased coefficients under specific
conditions. Appropriate instruments must (Murray, 2006): i) not be economically relevant in our
model of interest (second-stage); ii) be relevant or correlated with our potentially problematic
variable; and, iii) be valid or uncorrelated with the error term in our model of interest (1). Effort
is devoted to ensuring that our set of instruments is appropriate, strong and valid. Economic
arguments are required to assert the exogeneity of our instruments, whereas relevance and
validity can, in some cases, be tested. The second criterion for appropriate instruments states
that the instruments must be relevant or “strong”. Weak instruments do not eliminate (and may
exacerbate) the bias of least squares. By applying two-stage least squares, we assess relevance
by inspecting the excluded instruments in the first-stage model. Explicitly, we specify a null
hypothesis stating that the excluded instruments are irrelevant in the first-stage and then calculate
an F-statistic. Large F-statistics reassure us that our instruments are strong. The third criterion,
point iii), is referred to in this paper as instrument validity not exogeneity. This is to distinguish
it from the economic arguments needed to back the instruments’ exclusion, or exogeneity, in the
second-stage. Instrument validity implies that the excluded instruments are uncorrelated with the
error term in the second-stage model. We approach validity via two channels. First, instrument
validity is tested using a Sargan test on over-identifying restrictions in models that contain more
instruments than endogenous variables. The null hypothesis is that all of the instruments are
valid. A rejection of the null means that some of the instruments are invalid, but the test
provides no information on which if any instruments are valid. For two of our models, we
calculate Sargan statistics. We also present models that are exactly identified – i.e., there are
exactly as many instruments as troublesome variables. In these cases, economic arguments are
11
used to assuage concerns about instrument validity. Finally, we use Wu-Hausman tests to
determine whether anything is gained by using two-stage least squares. Instrumental variable
methods allow consistent estimation of coefficients when least squares is problematic. However,
they trade-off efficiency in the form of larger standard errors for unbiasedness. Wu-Hausman
statistics test whether the loss of efficiency is justified. Specifically, the null hypothesis of the
Wu-Hausman states that both the least squares and instrumental variables coefficients are
consistently estimated, but that the instrumental variables estimator is less efficient. A rejection
of the null implies that only the instrumental variables coefficients are consistent. Of course,
non-rejection implies that the least squares estimates are unbiased, consistent and more efficient
than the comparable instrumental variable models. This test yields valuable information for
when we select our preferred results.
We are most concerned about bias due to the endogeneity of the price of gasoline. As
such, we apply the approach of Davis and Kilian (2010). The market price of gasoline is
instrumented with provincial, inflation-adjusted excise taxes. Excise taxes comprise a significant
share of the final price of gasoline, so the link between tax changes and price movements is
obvious (i.e, excise taxes are a strong instrument for the price of gasoline). However, Davis and
Kilian argue that changes in tax legislation occur with a “considerable lag” (1197) and therefore
any variation in tax rates are exogenous to short-run changes in gasoline demand. Marion and
Muehlegger (2011) make a similar assertion in their analysis of the incidence of excise taxes.
We assume this argument is reasonable and maintain it.
Next, there is a high probability that the carbon tax is exogenous. Still we exploit unique
features of the policy design to formulate an instrumental variable strategy. Specifically, the
revenue-neutrality of the carbon tax generates two potential instruments. Personal and corporate
income taxes were both reduced in conjunction with the introduction of the policy. Therefore,
aggregate provincial personal and corporate income tax revenue should be negatively correlated
with the carbon tax. It is challenging to find plausible channels through which total provincial
income tax revenues may be linked to per capita gasoline demand, especially as total expected
government revenues remain constant (as described above, an explicit goal of the policy was to
maintain revenue neutrality). So, even despite arguments in favour of the exogeneity of the
carbon tax, we consider these instruments to be especially appealing.
12
Identifying our key parameters is of utmost importance, yet we are equally interested in
testing the impact of the carbon pricing policy relative to equal changes in the price of gasoline.
Formally we do this by performing F-tests on the equality of the coefficients of carbon tax and
tax-exclusive market price. Rejection of the null of coefficient equality implies that the market-
determined price elasticity is statistically distinguishable from the carbon tax elasticity.
Rejection of the null also allows us to claim that the carbon tax is either more or less salient than
the market-determined price of gasoline and that equivalent price changes produced distinct
demand responses. Our tables present heteroskedasticity and autocorrelation consistent (Newey-
West) standard errors.
Finally, we perform a series of robustness checks. We include potentially important
omitted variables such as after-tax income and the share of small and compact car sales by
province. We also inspect distinct sub-samples and rule out storage and announcement effects as
explanations for our findings.
4. ECONOMETRIC RESULTS
4.1 Overview of Main Findings
Three sets of results are presented. First, estimates from parsimonious least squares
models are discussed. Results from instrumental variable models and robustness checks are then
reviewed. Overall we find economically meaningful and statistically significant coefficients
when we use monthly data. Regressions using annual data are less precise and often have
confidence intervals that include zero. Point estimates from the annual models do buttress the
monthly results however.
Figure 2 displays results from our preferred model. The figure plots the coefficients from
a regression of monthly gasoline consumption on the interactions of the carbon tax and market
price of gasoline with the fixed effect for the province of BC (model (2) in Table 2). Also
presented are the 50% (the bold line) and 95% (thin line) confidence intervals. Clearly, the
carbon tax has a much larger impact on gasoline demand than do market prices. The point
estimate on the carbon tax equals -0.025 which, for a carbon tax of $25 tCO2e, implies a 12.5%
decrease in gasoline demand. An equivalent increase in the market price of gasoline predicts a
1.8% reduction in demand (the coefficient equals -0.004). Stated differently, the BC carbon tax
13
generated a demand response that is 7.1 times greater than an equivalent increase in market
prices.
Upon initial consideration, the sizeable difference between the elasticities for the carbon
tax and price of gasoline may surprise. However, there is convincing corroborating evidence
bolstering the external validity of the finding. Even though designed to be revenue-neutral, in
actuality, the BC carbon tax has been revenue negative. The carbon tax has collected less
revenue than the government initially forecast. Indeed, only 83% of the initially forecast
revenues has come in, well-below the benchmark of revenue-neutrality and a deficit too large to
ascribe to poor forecasting (BC Ministry of Finance, 2012b). Our main conclusion appears
sound: the carbon tax causes a larger change in demand than equivalent changes in market prices.
4.2 Least Squares Results
Table 2 presents results from the least squares models. Coefficient estimates and
standard errors for four models are presented in the upper section, while test results for the
salience null and model interpretation are at the bottom. Both specifications use monthly data,
include provincial and time fixed effects and have log-linear formulations.
Table 2 shows that the coefficient on the market price of gasoline in (1) is statistically
insignificant and equal to -0.0014. The coefficient of the carbon tax in this model is statistically
significantly different from zero and equals -0.0457. Weighing these against parameters in (2)
where prices and taxes are interacted with provincial fixed effects, several notable differences
surface. First, BC residents are more sensitive to changes in the price of gasoline compared with
the country as a whole. The absolute value of the estimate for price in (2) is nearly three times
larger than in (1). Similarly, comparing the coefficients on the carbon tax between the two
models reduces the initial estimate from -0.0457 to -0.0247, likely because in (1) the carbon tax
was capturing BC’s added responsiveness to gasoline prices.
Focusing on (2), a five cent increase in the price of gasoline leads to a 2.2% reduction in
the demand for gasoline in BC. Barla et al. (2009) estimate short-run price elasticities for
Canadian gasoline demand of roughly -0.1. With an average gasoline price of approximately 60
cents per litre during the period of the study, Barla et al.’s elasticities imply a 5 cent increase in
gasoline price reduces short run demand by about 0.8%. These estimates can be compared to (1),
which shows a five cent increase in the price of gasoline reduces demand by 0.7%. In general,
14
the results for the responsiveness of gasoline demand to changes in price are consistent with the
literature. Unlike previous studies however, we are interested in the relative responsiveness of
demand to carbon taxes and prices. Based on the F-test, we are able to reject the null hypothesis
of identical consumer responses to prices and carbon taxes in both (1) and (2). Carbon taxes are
statistically more salient than equivalent market prices.
The estimates that we focus on are in (2). We find that in the short-run a $25 tCO2e tax
yields a 12.5% reduction in the demand for gasoline, an effect that is 7.1 times greater than
would be expected from an equivalent increase in the market price of gasoline. There are three
reasons why this is our preferred model: i) the coefficients are precisely estimated, ii) the point
estimates are in the neighbourhood of the majority of the other specifications, and iii) the model
is simple and parsimonious. In section 5.2, this is the model used to calculate the counterfactual
scenarios and determine the reduction in tonnes CO2e emitted.
4.3 Instrumental Variables Results
Table 3 presents the coefficients from the instrumental variable models. Also displayed
are the diagnostics on instrument relevance and validity. The dependent variable in (1) is
monthly per capita gasoline consumption. (2) and (3) have annual per capita gasoline
consumption as the dependent variable. The bottom of the table interprets these coefficients and
presents the F-tests for the salience hypothesis.
In (1), the monthly market price of gasoline is instrumented with provincial excise tax.
The F-statistic on the excluded instruments in the first-stage is 435.52 indicating that, as
expected, excise taxes strongly predict gasoline prices. Testing whether the instrument is
correlated with the second-stage error term is not possible as the model is exactly identified.
Examining the coefficients, the point estimates for the price of gasoline and carbon tax equal
-0.0033 and -0.0257, respectively. These are comparable with (2) in Table 2. The instrumental
variables models are less precisely estimated and, as a consequence, we can only reject the null
hypothesis that the effect of the carbon tax is statistically different from zero at a 5% level of
significance (compared to 1% in Table 2). Yet, non-rejection of the null hypothesis in the Wu-
Hausman test indicates that the least squares estimates, those from (2) in Table 2, are unbiased,
consistent and efficient – i.e., the difference between the least squares and instrumental variables
15
estimates is purely random. Both the consistency of the coefficient estimates and the Wu-
Hausman statistic substantially bolster the credibility of our preferred model.
(2) and (3) in Table 3 use annual data as information for our carbon tax instruments is
only available on a yearly basis. The potentially problematic variable in (2) is the carbon tax.
Provincial personal and corporate income tax revenues are used as instruments. The F-statistic
on the excluded instruments in this case equals 13.36 indicating that the instruments are relevant.
Moreover, there are more instruments than endogenous variables, so a Sargan test is formulated.
The null hypothesis for the Sargan is that all of the instruments are valid. That we cannot reject
the null offers confidence that our instruments are valid in this formulation. The point estimate
for the carbon tax variable however is imprecisely estimated in (2) and equals -0.0054. It is not
statistically significantly different from zero. Further, even though the carbon tax elasticity is
larger than the market price elasticity, we are unable to reject the null of coefficient equality.
Still, as in (1), we are not able to reject the null hypothesis of the Wu-Hausman test for either (2)
or (3), indicating that equivalent least squares estimates would be more efficient than the
instrumental variables results.
Finally, (3) in Table 3 considers both the market price of gasoline and the carbon tax as
potentially problematic. These results further colour the findings from the monthly models. The
point estimates for the coefficients resemble the other models equalling -0.0015 and -0.0278 for
gasoline prices and the carbon tax. Yet, unlike the previous results, the null of the Sargan and
Wu-Hausman tests must be rejected, suggesting that this specification may be problematic.
Further, the F-statistic on the excluded instruments is smaller than in the other formulations. The
Angrist-Pischke multivariate F-statistic (Angrist and Pischke, 2009) for excise taxes equals 15.95,
while it is 6.55 for personal and corporate taxes. Despite the rejection the Sargan test, the point
estimates confirm the main conclusion and our preferred model uses monthly data. Thus, while
some concern is warranted when we consider both the price of gasoline and the carbon tax as
potentially endogenous within the same model, the implications of the instrument invalidity
appear to be minor.
The interpretation of the instrumental variable models is at the bottom of Table 3. For (1),
the model that uses monthly data and considers the price of gasoline to be endogenous, a five
cent increase in the price of gasoline then yields a 1.7% reduction in litres consumed. This
should be compared with the 1.8% decrease from (2) in Table 2. The same five cent increase in
16
the carbon tax reduces demand by 13.0% in the instrumental variable model compared with
12.5% using the least squares coefficients. However, while the monthly results support the
conclusions from Table 2, the annual formulations cloud interpretation. In particular, it is not
possible to reject the hypothesis that the carbon tax generated a statistically significantly
different demand response when compared to the market price of gasoline when we use the
lower resolution, annual data. Even though we cannot statistically claim that the carbon tax is
more salient than gasoline prices, we do not believe that jeopardizes our main interpretation. At
risk of belabouring the point, instrumental variable estimates have inflated standard errors
relative to least squares models and we cannot reject the null in the Wu-Hausman test. Therefore,
it is more appropriate to rely on the least squares models with the higher frequency data.
Nonetheless, the inability to fully uphold the finding in Table 2 must be noted. Using the point
estimates and a $25tCO2e carbon tax, the coefficients from (1) imply that the carbon tax
generates a demand response that is 7.9 times greater than would be expected from an equivalent
increase in the market price of gasoline. At $25 tCO2e, the carbon tax reduces gasoline demand
by 1.3 and 18.8 times more than equivalent price movements in (2) and (3), respectively, and a
five cent increase in the carbon tax leads to a corresponding 2.7% and 14.1% reduction in
gasoline demand. An equivalent increase in market prices only reduces gasoline consumption by
2.1% in (2) but by 0.8% in (3).
4.4 Robustness Checks
Table 4 displays results from a range of robustness checks. Ten models are presented.
Across all specifications, the coefficients for both the market price of gasoline and the carbon tax
are consistent with and corroborate the results from the least squares and instrumental variables
models. Taken together, these specifications reinforce the main conclusion that the carbon tax
policy generated a larger demand response than would be expected from an equivalent increase
in the market price of gasoline.
After-tax income is included in models in (1) and (2) of Table 4. Data on income are
only available on an annual basis for 1990-2010. For (1) then, monthly income is imputed by
dividing by twelve on a year-over-year basis. It is important that after-tax rather than gross
income is used: due to the tax shift (revenue-neutrality) personal income taxes were reduced in
conjunction with the introduction of a carbon tax, so individuals actually had more money to
17
allocate between consumption and savings.11 Including income produces almost no change in
the estimated response to the market price of gasoline with coefficients similar to those in Table
2. Demand responses to the carbon tax are tempered by the inclusion of income, but only by a
small amount. This is likely due to the lower level and shorter span for which the carbon tax is
active; still, the response to the carbon tax relative to the price of gasoline remains 5.5 times
larger. When we re-estimate (1) using first-differences rather than levels, similar results
materialize. (2) shows these first-differenced results and demonstrates that our conclusions are
not sensitive to the use of levels. Based on these specifications, income effects do not appear to
drive the larger consumer response to the carbon tax. All key parameters are roughly consistent
to the inclusion or exclusion of income and not unduly influenced by specification.
Our analysis uses real prices. It is possible however that consumers make decisions
based upon nominal carbon tax levels rather than adjusted rates – in effect, consumers may
incorrectly estimate the influence of inflation. As the carbon tax was introduced via a series of
discrete steps, we able to exploit this feature of the program design to explore this potential
explanation. In (3), we use the nominal carbon tax to instrument for the real level (identifying
off the jumps) and find coefficient estimates agree with our previous specifications – the
estimates for market price and carbon tax, respectively, are -0.0036 and -0.0239. As saliency
persists, the differential responsiveness to the two prices appears robust to consumer
misoptimization with respect to inflation.
Next, a range of chronological effects may also potentially impact our results. First,
temporary, temporally contingent effects around the date of the carbon tax introduction could
influence our estimates. For example, estimates may be measuring an announcement effect and
not a distinct behavioural response to equivalent price changes. There are two main ways that an
announcement effect may manifest itself: i) individuals may reduce their consumption in the
month or two months following the introduction of the tax because they are more attentive to
their level of gasoline consumption or ii) individuals may shift their consumption to earlier
months or purchase and then store fuel in the months preceding the carbon tax. With respect to
storage, Borenstein, Bushnell and Lewis (2004) and Marion and Muehlegger (2011) state that the
capacity gasoline storage is extremely limited. Selected purchasers may be able to store some
11 After‐tax income refers to income after federal and provincial income taxes only. It does not account for other taxes such as sales or excise.
18
fuel, but capacity generally does not extend for more than a month. Along the same lines, we
must distinguish between temporary and permanent announcement effects. An announcement
effect associated with the carbon tax that causes a permanent change in behaviour should be
interpreted differently when compared with an announcement effect that causes households to
alter their driving patterns for a brief period before reverting to old habits. The former situation
emphasizes the prospective saliency or information content of the carbon pricing policy, whereas
the latter may be considered a fad and the true effect is not persistent. In (4) and (5), the months
of June and July and May through August are, respectively, dropped from the dataset. If an
announcement effect is present, either due to a storage decision or temporary change in
behaviour, the coefficient on the carbon tax should be smaller and closer to the estimate for the
price of gasoline. We see the opposite. Even with the limited sample, estimates for both
variables remain consistent. Month-year time fixed effects are included in all of the models, so
perhaps this result is not surprising. All the same, this enables us to conclude that the saliency of
the carbon tax is not driven by an announcement effect, rescheduling of travel plans or storage.
Similarly, if a structural break in the demand for gasoline occurred prior to the carbon tax policy,
the coefficients may reflect a coincident change in preferences. Even though Hughes, Knittel
and Sperling (2008) demonstrated that the short-run demand for gasoline has become more
inelastic in recent decades, it is possible that demand in BC has become more elastic and that it is
this change in preferences that the carbon tax coefficient is capturing. In (6), we limit the
monthly sample to January 2007 through December 2011 and find no appreciable difference in
the coefficients. The behavioural response to the carbon tax is still 8.5 times larger at the mean
using the restricted sample.
Interpreting the salience effect requires us to assume that consumers formulate
expectations consistently. Anderson, Kellogg and Sallee (2011) find that consumers use a “no-
change” to forecast gasoline prices, so we would expect identical responsiveness to both taxes
and market prices.12 Nonetheless, it is possible that BC residents made errors with respect to
their expectation of carbon tax increases – namely, they may have behaved as though the tax was
initially set at its maximum $30 tCO2e level. To test this hypothesis, we artificially set the
carbon tax to its maximum nominal level throughout our period of analysis and then adjust for
12 Similarly, there should be identical responsiveness if consumers expect prices to be mean‐reverting or random walks.
19
inflation. Beyond exploring consumer misclassification of tax levels, this specification has
additional value as it provides an upper bound on the carbon tax elasticity. (7) displays the
results from this formulation: consumers, even in this extreme case, are still approximately three
times more responsive to changes in carbon taxes than they are to equivalent market price
fluctuations. The coefficient on gasoline prices matches other models at -0.0035, while the
estimate for the carbon tax is approximately half of our preferred specification.
Three final robustness checks are completed. In (8), provincial and month fixed effects
are interacted and a trend is added. With this model, we eliminate the possibility of province-
specific seasonality as an explanation for the added saliency of the carbon tax. Again the
estimated coefficients are in the expected neighbourhood. Finally, (9) and (10) include the
percentage of small and compact car sales out of all new car and all new vehicle sales as
potentially omitted variables. The coefficients on these variables have the appropriate sign, but
have little influence on the point estimates for gasoline prices and the carbon tax. In section 5.3,
we discuss several other non-testable explanations for our results. However, the robustness of
our key coefficients across a range of specifications and sub-samples strengthens our conclusion.
5. DISCUSSION
5.1 Intuition for Salience of Carbon Taxes
Rationalizing the saliency of the carbon tax is challenging within conventional consumer
theory. Equivalent price changes for the same good, whether motivated by policy or exogenous
supply shocks, should yield identical changes in quantity demanded. We argue that carbon tax
salience can be characterized as a resentment of free-ridership effect. In general, behavioural
economics has been applied to explain the saliency phenomenon. Finkelstein (2009), Chetty,
Looney and Kroft (2009) and Goldin and Homonoff (2010) focus on individual irrationality in
the form of inattentiveness or cognitive costs as a motive for their results. Consumers
misoptimize or apply rules-of-thumb and these errors manifest themselves as empirically distinct
responses to taxes and prices. Congdon, Kling and Mullainathan (2009) describe how other
insights from behavioural economics have yet to be explored within the tax policy literature. In
20
particular, they highlight non-standard preferences as an overlooked area.13 Non-standard
preferences refers to the notion that individuals may be “other-regarding” or not “perfectly self-
interested” or may use “reference points” (377). The welfare of others and, in this case, the
environment may enter directly into the utility function.
Intuitively, the notion of resentment of free-ridership can be characterized as a
manifestation of other-regarding preferences where carbon taxes enter the utility function both
through prices and as a direct argument because individuals care about their own and others’
contribution to an environmental public good. Building on Chetty (2009), assume a consumer
allocates her budget between two goods x and y. Good x, the numeraire, is a “clean” good and
untaxed. Consumption of good y, gasoline, generates CO2e emissions on which the government
levies a carbon tax t. The consumer has wealth Z and utility: tyvxu , . Let the exogenous
price of y be tpq , where p is the carbon tax-exclusive price of gasoline. At the optimum,
the consumer selects a consumption pair, ZtqyZqx ,,,, ** such that **, xuqtyv and
Zqyx ** . Focusing on the effect of the carbon tax on gasoline demand, differentiate
Ztqy ,,* with respect to t14: t
q
q
y
t
y
dt
dy
q
*
. Herein the carbon tax has two effects, a direct
effect on utility and a price effect. To start, observe that if 0
qt
y, then there is only a price
effect and the standard commodity taxation prediction is: t
y
p
y
**
(a result that is not
necessarily supported by empirical research (Finkelstein, 2009; Chetty, Looney, Kroft, 2009)).
Most research on carbon taxation focuses the exclusively on this price effect. With other-
regarding preferences however, tax policy may act as a focal point or coordinating mechanism
leading to individual behavioural change. This phenomenon, which is especially relevant for
environmental public goods where coordination is required to appropriately overcome the
influence of negative externalities, would be observable via the consumer’s demand schedule.
13 While less precise with respect to the mechanism, (e.g., learning, statues, altruism, resentment, etc.), the burgeoning peer effects literature in environmental economics (e.g., Allcott (2011) and Bollinger and Gillingham (2012)) can be categorized as other‐regarding preferences. 14 Income is suppressed.
21
Consider the case of resentment of free-ridership as a type of other-regarding preference
and explanation for why this shift occurs. Assume that an environmentally conscious driver
wants to contribute to an environmental public good (i.e., lower her carbon emissions) by
reducing the amount that she drives.15 Without a price on carbon, one outcome of her decision to
drive fewer kilometres is that it lowers the cost of driving for the non-environmentally conscious
drivers, enabling them to drive more. This is a form of leakage where actual emission reductions
from the environmentally conscious driver are eliminated by increases in emissions from other
drivers. Even if there are net positive environmental benefits, the environmentally conscious
driver may be resentful of this leakage. This resentment may manifest itself as a disincentive to
contribute to the public good. Ultimately, resentment of free-ridership may cause the
environmentally conscious driver to contribute less to the public good than if free-riding was not
possible. Imposition of a carbon tax eliminates the free-ridership problem and, as a consequence,
any resentment of free-ridership. A carbon tax forces all drivers to pay an environmental cost for
each litre of gasoline consumed. An environmentally conscious driver can therefore reduce her
kilometres knowing that the non-environmentally conscious individual is paying the full
environmental costs of his decision. In this example, the driver has non-standard preferences
with respect to the carbon tax and the existence of the tax itself causes a change in behaviour
beyond any price effects.
Resentment of free-ridership predicts 0
qt
y. Yet, a converse argument could be made
for a relief of guilt effect. Herein, an individual may have driven less prior to the introduction of
the carbon tax because she felt guilty about the environmental damage she generated. Once she
knows that she is able to pay the full environmental costs of her driving decisions – i.e., once a
carbon tax exists – she is willing to drive more implying 0
qt
y. A relief of guilt effect would
mitigate a tax’s impact, leading to the conclusion that the tax is less salient than equivalent price
15 In return, the environmentally conscious driver may, for example, receive a warm glow (Andreoni, 1990; Kotchen, 2006). It worth noting that having the tax as an argument in the utility function means that its effect is distinct from that predicted by a warm glow or impure public good model (e.g., Cornes and Sandler, 1994; Viscary, 2011; Kotchen, 2007). In impure public good models, taxes still work through prices. Nonetheless, under certain conditions, the impure public good model does predict empirically observable features of the gasoline market (e.g., asymmetric responses to price increases and decreases).
22
changes. Ultimately, under this simple other-regarding preferences logic, the saliency of
environmental taxes and whether a resentment of free-ridership or a relief of guilt effects exist is
an empirical question.
5.2 Reductions in Quantity of Gasoline Demanded and Carbon Emissions
Three scenarios are constructed to determine the reduction in litres demanded and emitted
tonnes of CO2e attributable to the BC carbon tax.16 We use the coefficients from our preferred
model throughout. Scenario one is a baseline counterfactual which represents the situation had
no carbon policy been introduced. Scenario two represents a situation where it is assumed that
the carbon tax has an elasticity of demand which is identical to the price of gasoline – i.e., it is
the “equivalent response” scenario. Lastly, scenario three is the “salient carbon tax” scenario
and employs the model-predicted quantity demanded.
Figure 3 plots for 2007 through 2011 the annual average of monthly gasoline demand for
the three scenarios. Scenario one reflects the (simulated) baseline counterfactual demand for
gasoline. It charts the number of per capita litres gasoline consumed per month without any
carbon price. Demand remains relatively stable with fewer litres demanded in 2008 than in 2011.
Scenario two has a corresponding shape to scenario one; however, a growing wedge, reflecting
the carbon taxes’ increase from 2.3 cents per litre in 2008 to 5.6 cents per litre in 2011, is
apparent. Scenario two reflects a counterfactual where the carbon tax had the same influence as
any other increase in the price of gasoline. Finally, scenario three shows the decrease in quantity
of gasoline demanded that is the result of the differential response to the carbon tax. The
elasticity of demand for the carbon tax is nearly five times larger than for market prices.
Obviously, this translates into a larger decline in litres demanded which is exhibited in the figure.
In 2011, predicted per capita demand per month is 13.5 litres lower in scenario three than in
scenario one. In fact, demand in scenario three continues to trend downward even as scenarios
one and two increase.
Figure 4 illustrates the province-wide, monthly reduction in tCO2e attributable to the
carbon tax. For each scenario, predicted per capita monthly litres demanded are multiplied by
population to obtain the aggregate monthly demand of gasoline for the entire province of BC.
16 Note that the BC carbon tax affects all fossil fuels consumed in the province, not just gasoline. However, our analysis is restricted to the impact of the carbon tax on gasoline sales.
23
Volume of fuel demanded is then converted into tCO2e to obtain the gross tonnes emitted under
the three scenarios.17 Using scenario one as the baseline, we calculate the reduction in emissions
from a situation where no carbon price policy is enacted. The dark bars depict the difference
between scenarios one and two. This is the saved emissions had the carbon tax yielded the same
demand response as conventional price increases. The lighter grey bars reflect the actual
emission reduction. Over the first four years, the BC carbon tax reduced emissions from
gasoline consumption by 3.61 MtCO2e. The large majority (about 85%) of this reduction is due
to the saliency of the tax.
5.3 Other Potential Explanations for the Results
Two alternative explanations, cross-border shopping and contemporaneous vehicle
efficiency policies, could also potentially explain our findings. Both are discussed in turn. We
believe that neither is credible and think that there are good reasons to trust that it is the carbon
tax that caused the change in consumer behaviour.
The first alternative explanation is that the carbon tax coefficient is actually capturing a
cross-border shopping effect. Higher prices due to the carbon tax may have encouraged drivers
to begin filling their gas tanks in neighbouring jurisdictions. We are unable to dismiss this
explanation, yet believe it to implausible. Alberta is to the east of BC, while Washington State is
to the south. Both jurisdictions have had lower gasoline prices than BC for many years,
including a lengthy period prior to the introduction to the carbon tax (refer to Panel B in Figure
1). Further, over 95% of BC’s population would need to drive more than two hours to reach the
Alberta border, while crossing into Washington involves an international border. It is doubtful
that a sizeable share of residents suddenly began cross-border shopping because of the carbon tax.
Nonetheless, this is a potential alternative explanation for our findings.
BC also increased funding for its accelerated vehicle retirement program, known as
SCRAP-IT, in the two years following the introduction of the carbon tax (Antweiler and Gulati,
2011). Likewise it offered subsidies for hybrid vehicle purchases in the years preceding the tax
17 Each litre of gasoline combusted generates 2.47kgCO2e. This assumes that the energy content of gasoline is 35MJ/L, while the CO2e content is 70.68 tCO2e/TJ. Putting it together: 1L = 35MJ * 0.07068 kg CO2e/MJ = 2.4738kgCO2e. Note that the BC government uses a slightly different carbon content adjustment because of the exemption for the biofuel mandate that exists in the province. Our calculations effectively assume that over their lifecycle biofuels generate the same emissions as conventional gasoline.
24
(Chandra, Gulati and Kandlikar, 2010). Both policies can be considered coincident with the
carbon tax. One concern is that these programs sufficiently altered the composition of the
vehicle stock, increasing the average fuel economy and leading us to misattribute the reduction
in gasoline demanded to the carbon tax. Several arguments lead us to believe this is unlikely.
The main reason is the scope of the programs was too small. SCRAP-IT, a program which
offered individuals incentives to purchase more fuel efficient cars18 covered 18,000 vehicles over
2008 to 2011 period (Antweiler and Gulati, 2011). This is less than 1% of the over 2 million
passenger vehicles registered in the province (BC Statistics, 2011).19 Even if every vehicle that
entered the SCRAP-IT program was removed from the road, the total litres of gasoline saved
would account for only a small fraction of the decrease in gasoline demand. Besides, the
majority of SCRAP-IT vehicles were replaced with newer cars and our robustness checks
demonstrate that the carbon tax coefficient is not sensitive to small and compact car sales. Along
the same lines, hybrid vehicles comprised only 1.1% of all vehicles sold in BC in 2006 and
Chandra, Gulati and Kandlikar (2010) demonstrate that hybrid vehicles tend to replace other
small or fuel efficient vehicles. So, the actual increase in fuel economy is small. On balance,
these programs are simply too minor to jeopardize our main conclusions. Moreover, BC
residents historically tended to purchase vehicles that were already more fuel efficient than
average (Chandra, Gulati and Kandlikar, 2010), implying that any accelerated retirement or
hybrid subsidy program would have less of an effect in BC than might be found in other
provinces or states.
We emphasize that our results are limited to the short-run, so explanations based on a
durable goods model of consumer behaviour are not appropriate. For example, while it is
possible that risk averse consumers make different vehicle purchase decisions when faced with a
fixed carbon price compared to a volatile gasoline price. Our analysis is based primarily on
monthly data where very little turnover in vehicle stock is present. Additionally, in selected
specifications, we control for characteristics of the vehicle stock without a substantive effect on
coefficients.
18 Participants could also switch to public transit or receive a subsidy in order to purchase a bicycle (Antweiler and Gulati, 2011). 19 There are also the more than 700,000 commercial vehicles registered in the province (BC Statistics, 2011).
25
As a final point, it is important to note that there is strong evidence for the external
validity of our findings. As mentioned, the carbon tax was designed to be revenue-neutral but is,
in fact, revenue-negative. The carbon tax will collect $510 million less than expected over its
first four years.20 This is a 16.7% shortfall, an amount too large to blame on forecasting error.
The carbon tax does seem to have generated a larger demand response than anticipated.
6. CONCLUSION
This paper presents the first rigorous empirical evaluation of an actual carbon tax within a
North American context. Through a wide-range of econometric specifications, we demonstrate
that the carbon tax introduced by the Canadian province of BC is more salient than equivalent
changes in price. A five cent increase in the carbon tax, all else constant, causes gasoline
demand to decline by 12.5% whereas an identical five cent increase in the market price of
gasoline leads to a 1.8% reduction in litres consumed. At $25/tCO2e, the carbon tax is 7.1 times
more salient than the market price of gasoline. Finally, over the first four years of the policy, the
BC carbon tax led to a total reduction in emissions from gasoline consumption of over 3.5
million tCO2e when compared with a counterfactual scenario of no tax.
To our knowledge, this is the first analysis of the saliency of environmental pricing. The
results contribute to a prevailing shift to incorporate behavioural economics into tax and
environmental policy. While current theory points to individual irrationality as an underlying
mechanism for the saliency phenomenon, the discussion in this paper signals that individuals
have may also have non-standard preferences – in this case with respect to environmental public
goods. As the tax saliency literature matures, an improved understanding of how markets and
policy interact within consumer utility functions will develop. We view this research as a
contribution to both the behavioural economics of tax policy as well as providing rigorous
empirical estimates for the demand elasticity of an actual carbon tax.
Our findings have several direct environmental policy implications. Most obvious is that
individuals do not necessarily respond in the same way to tax increases as to supply shocks.
Although our analysis is based on a single policy, the results should be taken into account by
policy-makers who are considering introducing environmental taxes which affect gasoline prices.
20 This is an undiscounted sum of the difference between forecast and actual revenues over the first three years plus the difference between the initial forecast and updated forecast for the most recent fiscal year.
26
Indiscriminate use of gasoline elasticities may generate inaccurate forecasts of tax revenue and
emissions. Our analysis also adds another wrinkle to the price uncertainty versus quantity
uncertainty debate that has coloured discussions of tax versus cap-and-trade systems. In
particular, this research suggests that environmental taxes that directly affect consumer prices
result in a larger demand response than an equivalent supply shock. Depending on how
consumers view allowance prices (as an environmental ‘price’ or as a supply shock) consumers
may respond quite differently to equivalent prices in cap and trade and carbon tax policies,
adding another dimension to the comparison of these instruments.
27
7. REFERENCES
Allcott, H. 2011. “Social Norms and Energy Conservation.” Journal of Public Economics, 95: 1082-1095. Anderson, M.S., 2010. “Europe’s Experience with Carbon-energy Taxation.” S.A.P.I.E.N.S., 3(2): 1-11. Anderson, S.T., R. Kellogg and J.M Sallee. 2011. “What do Consumers Believe about Future Gasoline Prices?” NBER working paper, no. 16974. Andreoni, J. 1990. “Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving.” Economic Journal, 100(401): 464–477. Angrist, J.D. and J-S Pischke. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press: Princeton. (Correction posted at: www.mostlyharmlesseconometrics.com/2009/10/multivariate-first-stage-f-not/) Antweiler, W. and S. Gulati. 2011. “An Analysis of British Columbia’s SCRAP-IT Program: Emissions Savings, Participation, and Transportation Choice.” Paper presented at European Association of Environmental and Resource Economists’ 18th Annual Conference, 29 June - 2 July 2011, Rome. (http://www.webmeets.com/EAERE/2011/m/viewpaper.asp?pid=1396.) Barla, P., B. Lermonde, L.F. Miranda-Moreno and N. Boucher. 2009. “Traveled Distance, Stock and Fuel Efficiency of Private Vehicles in Canada: Price Elasticities and Rebound Effect.” Transportation, 36: 389-402. Bollinger, B. and K. Gillingham. 2012. “Peer Effects in the Diffusion of Solar Photovoltaic Panels.” Marketing Science, 31(6): 900-912. Borenstein, S., J. Bushnell and M. Lewis. 2004. “Market Power in the California Gasoline Market,” Center for the Study of Energy Markets, working paper 132. British Columbia Ministry of Finance. 2011. “Consolidated Revenue Fund Extracts.” http://www.fin.gov.bc.ca/ocg/pa/10_11/PA_2011_CRF_Extracts.pdf, (last accessed June 15, 2012). British Columbia Ministry of Finance. 2012a. “Myths and Facts about the Carbon Tax.” http://www.fin.gov.bc.ca/tbs/tp/climate/A6.htm, (last accessed April 17, 2012). British Columbia Ministry of Finance. 2012b. Budget and Fiscal Plan: 2012/13-2014/15. http://www.bcbudget.gov.bc.ca/2012/bfp/2012_Budget_Fiscal_Plan.pdf. British Columbia Statistics. 2011. “Other Economic Statistics: Licenced Motor Vehicles.” http://www.bcstats.gov.bc.ca/StatisticsBySubject/Economy/OtherEconomicStatistics.aspx. (Last accessed July 17, 2012).
28
Bruvoll, A. and B.M. Larsen. 2004. “Greenhouse Gas Emissions in Norway: Do Carbon Taxes Work?” Energy Policy, 32: 493-505. Chandra, A., S. Gulati and M. Kandlikar. 2010. “Green Drivers or Free Riders? An Analysis of Tax Rebates for Hybrid Vehicles.” Journal of Environmental Economics and Management, 60(2): 78-93. Chetty, R. 2009. “The Simple Economics of Salience and Taxation.” NBER working paper, no. 15246. Chetty, R., A. Looney and K. Kroft. 2009. “Salience and Taxation: Theory and Evidence.” American Economics Review, 99(4): 1145-1177. Congdon, W.J., J.R. Kling and S. Mullainathan. 2009. “Behavioral Economics and Tax Policy,” National Tax Journal, LXII(3): 375-386. Cornes, R. And T. Sandler. 1994. “The Comparative Static Properties of the Impure Public Good Model.” Journal of Public Economics, 54: 403-421. Davis, L.W. and L. Kilian. 2011. “Estimating the Effect of a Gasoline Tax on Carbon Emissions.” Journal of Applied Econometrics, 26: 1187-1214. DesRosiers Automotive Consultants Inc. 2012. Vehicle Sales. http://www.desrosiers.ca/index.html. Enevoldsen, M.K., A.V. Ryelund and M.S. Andersen. 2007. “Decoupling of Industrial Energy Consumption and CO2-Emissions in Energy-intensive Industries in Scandinavia.” Energy Economics, 29(4): 665-692. Finkelstein, A. 2009. “E-ztax: Tax Salience and Tax Rates.” Quarterly Journal of Economics, 124(3): 969-1010. Gallagher, K.S. and E. Muehlegger. 2011. “Giving Green to Get Green? Incentives and Consumer Adoption of Hybrid Vehicle Technology.” Journal of Environmental Economics and Management, 61(1): 1-15. Goldin, J. and T. Homonoff. 2010. “Smoke Gets in Your Eyes: Cigarette Tax Salience and Regressivity,” Working Papers (Princeton University, Industrial Relations Section), no. 561. Goulder, L. 1995. “Environmental taxation and the double dividend: A reader’s guide.” International Tax and Public Finance, 2(2): 157-183. Harrison, K. 2012. “A Tale of Two Taxes: The Fate of Environmental Tax Reform in Canada.” Review of Policy Research, 29(3): 383-407.
29
Hughes, J.E., C.R. Knittel and D. Sperling. 2008. “Evidence of a Shift in the Short-Run Price Elasticity of Gasoline Demand.” Energy Journal, 29(1): 93-114. Kent Marketing Services. 2012. Petroleum Price Data. http://www.kentmarketingservices.com/dnn/PetroleumPriceData.aspx. Kotchen, M.J. 2006. “Green Markets and Private Provision of Public Goods.” Journal of Political Economy, 114(4): 816-834. Kotchen, M.J. 2007. “Equilibrium Existence and Uniqueness in Impure Public Good Models.” Economic Letters, 97: 91-96. Kruger, J., W.E. Oates and W.A. Pizer. 2007. “Decentralization in the EU Emissions Trading Scheme and Lessons for Global Policy.” Review of Environmental Economics and Policy, 1(1): 112-133. Li, S., J. Linn and E. Muehlegger. 2012. “Gasoline Taxes and Consumer Behaviour.” NBER working paper, no. 17891. Marion, J. and E. Muehlegger. 2011. “Fuel Tax Incidence and Supply Conditions,” Journal of Public Economics, 95: 1202-1212. Martin, R., L.B. de Preux and U.J. Wagner. 2011. “The Impacts of the Climate Change Levy on Manufacturing: Evidence from Microdata.” NBER working paper, no. 17446. Murray, M.P. 2006. “Avoiding Invalid Instruments and Coping with Weak Instruments.” Journal of Economic Perspectives, 20(4): 111-132. Regional Greenhouse Gas Initiative (RGGI). 2012. “Welcome.” www.rggi.org, (last accessed April 17, 2012). Small, K.A. and K. Van Dender. 2007. “Fuel Efficiency and Motor Vehicle Travel: The Declining Rebound Effect.” Energy Journal, 28(1): 25-51. Statistics Canada. 2012. Cansim Database. http://www5.statcan.gc.ca/cansim/home-accueil?lang=eng. Sumner, J., L. Bird and H. Smith. December 2009. “Carbon Taxes: A Review of Experience and Policy Design Considerations.” National Renewable Energy Laboratory, Technical Report NREL/TP-6A2-47312. Vicary, S. 2011. “Public Goods and the Commons: A Common Framework.” Journal of Public Economic Theory, 13(1): 47-69. Yatchew, A. and J.A. No. 2001. “Household Gasoline Demand in Canada.” Econometrica, 69(6): 1697-1709.
8. FIGU
Figu
URES
ure 1: AveraGasoline C
age MonthlyConsumptio
y Tax-incluson (bottom
30
sive Gasolinpanel) in Br
ne Price (topritish Colum
p panel) andmbia and O
d per CapitOntario
a
Figure
e 2: Effect off Gasoline PEst
Prices and Ctimates and
31
Carbon Taxd Confidence
xes on Gasole Intervals
line Demand in BC: Po
oint
Figur
re 3: Averagge Monthly Col
per Capita lumbia: Cou
32
Consumptiunterfactua
ion of Gasolal Scenarios
line Consum
med in Briti
ish
Figuree 4: Reduction in Emiss
33
sions Attribbutable to thhe Carbon T
Tax
34
9. TABLES
Table 1: Key Characteristics of the Revenue Neutrality of the BC Carbon Tax Design*
Carbon Tax
($/tonnes)
Carbon Tax
(cents/litre)
Average
Provincial
Personal Income
Tax Rate on
$100,000
Provincial
Business Limit
Provincial
Corporate
Income Tax Rate
(high income)
Provincial Small
Business Tax
Rate
(low income)
July 1, 2007 0 0 8.74% $400,000 12.0% 4.5%
July 1, 2008 10 2.34 8.02% $400,000 11.0% 3.5%
July 1, 2009 15 3.33 7.89% $400,000 11.0% 2.5%
July 1, 2010 20 4.45 7.86% $500,000 10.5% 2.5%
July 1, 2011 25 5.56 7.83% $500,000 10.0% 2.5%
July 1, 2012 30 6.67 7.72% $500,000 10.0% 2.5%
* All non‐carbon tax rate changes enacted on January 1st.
In column 2, $/tonne refers to the price in Canadian dollars per carbon dioxide equivalent tonne. Column 3 displays the tax
in cents per litre of unleaded liquid gasoline as calculated by the BC Ministry of Finance. The Personal Income Tax rates
displayed in Column 4 are the average provincial tax rate for a household earning a nominal income of $100,000 per year up
to the point of the tax change (i.e., the tax rate is calculated such that all income is assumed to be earned instantaneously
on July 1st ). The provincial business limit in Column 5 is the level at which the high income corporate tax rate becomes
effective. Column 6 presents the corporate tax rate for business profits that are greater than the provincial business limit,
which is displayed in in Column 5. The Small Business Tax rate as shown in Column 7 applies to net income that is less than
the small business limit (Column 5).
35
Dependent Variable
Market Price of Gasoline
Market Price of Gasoline*BC
Carbon Tax
Carbon Tax*BC
Provincial Fixed Effects
Month‐Year Fixed Effects
Year Fixed Effects
Specification
Observations
F‐test ‐ H0: Market Price = Carbon Tax
F‐test ‐ H0: Market Price*BC = Carbon Tax*BC
Estimated percent reduction in per capita gasoline consumption caused by a 5 cent increase in the:
Market Price of Gasoline
Carbon Tax
*** ‐ significant at a 1% level; ** ‐ significant at a 5% level; * ‐ significant at a 10% level
Table 2: Least Squares Estimates of the Effect of Gasoline Prices and Carbon Taxes on Gasoline
2580
Monthly per Capita Gasoline Consumption
(litres)
(2)
‐0.0035***
(0.0009)
‐0.0247***
(0.0074)
(0.0014)
‐0.0457***
(0.0106)
(1)
Yes Yes
Yes
‐0.0014
Table 2 reports the least squares estimates of the percent change in the per capita number of litres
gasoline consumed that results from a corresponding one cent increase in the market price of
gasoline and the carbon tax. All models include provincial and time fixed effects (year‐month) and
the values in parentheses are the heteroskedasticity and autocorrelation consistent standard errors.
(1) reports national coefficients for the market price of gasoline and the carbon tax. (2) reports the
interactions for the province of British Columbia only. The null hypothesis of equality of the market
price and carbon tax coefficients is tested using an F‐test. The null is rejected in both models. Also
reported are the estimated percent reductions in per capita litres of gasoline consumed that would
occur with a five cent increase in the market price and carbon tax respectively. Finally, a ratio of the
relative influence of the carbon tax to market‐induced price changes is calculated.
7.1
0.7%
33.4
23.4%
1.8%
12.5%
Ratio of the demand response to the carbon tax relative to an equivalent change in market prices at
$25 per tCO 2e:
23.07***
log‐linear
2580
Yes
log‐linear
95.42***
36
Table 3: Instrumental Variable Estimates for the Effect of Gasoline Prices and Carbon Taxes on Gasoline Consumption
Dependent Variable
Market Price of Gasoline*BC
Carbon Tax*BC
Provincial Fixed Effects
Month‐Year Fixed Effects
Year Fixed Effects
Specification
F‐statistic for Excluded Instruments(a)
Excise Taxes
PIT and CIT
Wu‐Hausman Test
(p‐value)
Sargan Test
(p‐value)
Observations
Estimated percent reduction in per capita gasoline consumption caused by a 5 cent increase in the:
Market Price of Gasoline
Carbon Tax
Ratio of the demand response to the carbon tax relative to an equivalent change in market prices at $25 per tCO 2e:
*** ‐ significant at a 1% level; ** ‐ significant at a 5% level; * ‐ significant at a 10% level
(a) ‐ The values reported in (3) are the Angrist‐Pischke multivariate F‐statistics (Angrist and Pischke, 2009).
F‐test ‐ H0: Market Price*BC = Carbon Tax*BC 6.38**
Annual per Capita Gasoline Consumption (litres)
Potentially Endogenous Variable
Instruments
(0.0082)
Yes
Yes
log‐linear
Market Price of Gasoline*BC
Provincial Gasoline Excise
Taxes
2580 216
(0.00)
Provincial Corporate Income
Tax Revenues
(1)
Monthly per Capita Gasoline
Consumption (litres)
‐0.0033***
(0.0012)
‐0.0257***
log‐linear
Market Price of Gasoline*BC
Carbon Tax*BC
Provincial Gasoline Excise
Taxes
Provincial Personal Income
Tax Revenues
0.12
(0.73)
Yes
log‐linear
Carbon Tax*BC
Provincial Personal Income
Tax Revenues
Provincial Corporate Income
Tax Revenues
435.52
13.36
0.19
(0.73)
0.00 0.98
(3)
‐0.0015**
(0.0007)
‐0.0041**
(0.0014)
(2)
216
‐0.0054
(0.0227)
Yes
‐0.0278***
(0.0078)
Yes
Yes
0.00
(0.98)
15.95
6.55
0.91
(0.99)
27.00
0.8%
14.1%
18.8
Table 3 displays the instrumental variable estimates for three models. In (1), the dependent variable is monthly per capita gasoline
consumption. (2) and (3) have annual per capita gasoline consumption as the dependent variable. For (1), inflation‐adjusted provincial
excise taxes are used as instruments. (2) intruments the carbon tax with the log of aggregate provincial personal and corporate income tax
revenue. (3) considers both market price of gasoline and the carbon tax as potentially endogeneous and uses the provicial excise tax and
personal and corporate income taxes as instruments. Values in parentheses are heteroskedasticity and autocorrelation corrected standard
errors. Also reported are the F‐statistics for the excluded instruments, the Wu‐Hausman test for endogeneity and the Sargan test of the
overidentifying restrictions. The bottom panel presents F‐tests for the equality of market price and carbon tax coefficients. Also reported
are the estimated percent reductions in per capita litres of gasoline consumed that would occur with a 5 cent increase in the market price
and carbon tax respectively.
7.9 1.3
13.0%
2.1%
2.7%
1.7%
37
(1) (2) (3) (4) (5)
Dependent Variable
Market Price of Gasoline*BC ‐0.0035*** ‐0.0017** ‐0.0036** ‐0.0034*** ‐0.0034***
(0.0010) (0.0004) (0.0011) (0.0009) (0.0010)
Carbon Tax*BC ‐0.0192** ‐0.0216** ‐0.0239* ‐0.0243** ‐0.0270***
(0.0093) (0.0004) (0.0144) (0.0076) (0.0079)
Income(a)
0.0164 0.0063
(0.0140) (0.0066)
Provincial Fixed Effects Yes Yes Yes Yes Yes
Month‐Year Fixed Effects Yes Yes Yes Yes Yes
Sample Restriction None None None Drop June‐July Drop May‐August
Specification log‐linear log‐linear log‐linear log‐linear log‐linear
None First DifferencesInstrument Real Tax
with NominalNone None
Observations 2460 2460 2580 2150 1730
(6) (7) (8) (9) (10)
Dependent Variable
Market Price of Gasoline*BC ‐0.0020 ‐0.0035*** ‐0.0030 ‐0.0025
(0.0019) (0.0009) (0.0019) (0.0019)
Carbon Tax*BC ‐0.0170*** ‐0.0130** ‐0.0319*** ‐0.0283***
(0.0065) (0.0051) (0.0121) (0.0104)
Market Price of Gasoline ‐0.0002
(0.0002)
Carbon Tax ‐0.0456**
(0.0021)
%age small cars out of car sales ‐0.7048
(0.5824)
%age small cars out of all vehicle sales ‐1.1806*
(0.6983)
Provincial Fixed Effects Yes Yes Yes Yes Yes
Month‐Year Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
Province*Month Fixed Effects Yes
Trend Yes
Sample Restriction2007‐2011 only
Set Carbon Tax at
Max Nominal Value NoneNone None
Specification log‐linear log‐linear log‐linear log‐linear log‐linear
Observations 600 2580 2580 216 216
** ‐ significant at a 1% level; * ‐ significant at a 5% level
(a)‐ Income is scaled by 100. Income data are available only until 2010.
Table 4: Effect of Gasoline Prices and Carbon Taxes on Gasoline Consumption: Robustness Checks
Table 4 presents a series of robustness checks. (1) includes after‐tax income into the model. (2) estimates the same model using first‐differences.
Column (3) instruments the real carbon tax with the nominal carbon tax, identifying the parameter off discrete jumps in the tax levels. (4) and (5) drop
June and July and May through August (resp.) to ensure that the coefficients on the market price of gasoline and carbon tax are not driven by
announcement or storage effects. A shorter time period, from 2007‐2011, is estimated in (6) to ensure that the results are robust to potential structural
changes. Column (7) sets the nominal carbon tax at the maximum value in an effort to test whether consumers behaved as though this rate were
effective throughout the period of analysis. This estimate provides an upper bound on the carbon tax elasticity. Provincial dummies are interacted with
month dummies and a time trend is included in (8). (9) and (10) include the share of small and compact cars sold as a percentage of all cars and vehicles
sold, respectively, in each province. Value in parentheses are heteroskedasticity and autocorrelation robust standard errors.
Monthly per Capita Gasoline Consumption (litres)
Monthly per Capita Gasoline Consumption (litres)Annual per Capita Gasoline Consumption
(litres)
Transformation of Variables
38
Mean StdDev Min Max Mean StdDev Min Max
Alberta 177.79 15.50 142.13 260.84 68.52 11.95 49.51 107.22
British Columbia 114.94 11.63 88.76 146.70 78.28 18.83 48.02 130.47
Manitoba 138.70 10.98 105.98 171.55 72.89 14.07 50.92 117.91
New Brunswick 143.99 17.34 106.81 203.15 77.33 14.28 54.99 119.41
Newfoundland 119.31 18.61 83.63 195.84 82.32 13.93 60.94 126.23
Nova Scotia 133.36 13.60 98.24 167.71 77.77 14.87 54.19 120.71
Ontario 128.47 7.78 109.88 149.87 72.84 13.82 53.35 114.68
Prince Edward Island 164.32 25.73 115.86 235.60 75.68 12.66 52.98 115.00
Quebec 112.89 7.70 89.36 140.64 79.19 14.84 56.55 126.40
Saskatchewan 205.48 35.92 132.73 303.03 75.80 13.42 49.25 117.26
Table A1: Summary Statistics for the Monthly Gasoline Consumption and the Market Price of
Gasoline consumption is in litres. Prices are inflation‐adjusted and tax‐inclusive.
Per Capita Gasoline Consumption Price of Gasoline
Top Related